Scale AI and Hugging Face have introduced a new vision-language-action framework called FineART-VLA, designed to help robots execute long, multi-step tasks by reasoning over actions sequentially. The release represents the first time Scale AI has open-sourced its robotics technology, aiming to provide developers with a ‘System 1-System 2’ model that can better manage robotic maneuvers.
The model is paired with the FineART dataset, which contains more than 40,000 demonstrations and 500,000 subtask labels across 151 distinct tasks. Both the reasoning model and the data library have been made available on the LeRobot platform to support the development of open-source robotics.
